Att skapa syfte för dungeons
Att utveckla processuellt narrativ för EDD med hjälp av macro patterns
To make sense of dungeons
Developing procedural narrative for EDD with the help of macro patterns
Alexander Flodhag
&
Simon Tolinsson
Huvudområde: Datavetenskap Examensarbetets nivå: Kandidatnivå Högskolepoäng: 15 hp
År, termin: 3, 6
Abstract
Together with the growth of procedural content generation in game development, there is a need for a viable generation method of procedural context to make sense of the content within game space. Previous research discusses how interactivity and narrative are almost opposite of each other and when combined needs to be generated in two steps, one for the game space and the other for context. We propose procedural narrative as context through objectives, as a useful means to structure content in games. In this paper we present and describe an artefact developed as a sub-system to the Evolutionary Dungeon Designer (EDD), that procedurally generates objectives for the dungeons created with the tool. The artefact is developed with macro patterns which can be defined as an extension of the existing meso patterns in EDD. The macro patterns are used to generate objectives in the rooms of the dungeons, and the system evaluates the priority of the room objectives based on the design of the dungeon and the quality of the objective to maximize the usage of game space and create a suitable narrative. The work for this thesis and its artefact resulted in a successful expansion on the knowledge of procedural narrative generation by presenting macro patterns as a viable solution for contextualization of procedural game content.
1. Introduction
Procedural content generation (PCG) has found itself in the spotlight within game development lately with games such as Minecraft, No Man’s Sky and Spelunky [1]. One of the reasons is because it can be used in many ways to benefit video games and its development process. One of the most common benefits is improving replayability, but also reducing the workload of the developers and fostering creativity for the designers [1, 2, 3]. But with this solution, other issues and challenges arise. Some of the challenges with procedural content in video games are e.g. creating non-generic and original content, overcoming the animation bottleneck and integrating music and other types of content [4]. However, this thesis is going to focus on the implementation of procedural narrative. Narrative is a form of context, which is a necessity when generated content is implemented into the game space since it is needed to make sense of the content [5]. If there is a lack of context in the generated content, then the user experience is going to be negatively affected and the content may end up being perceived as empty or meaningless [5]. An example of narrative is objectives. If there is an objective within the content, for example finding a sacred gem in a dangerous dungeon, then that creates interaction between the user and the content, which creates the needed context [5].
The motivation with this research is mainly to expand on the knowledge regarding procedural narrative, its generation, and to make sense of content by developing an artefact that creates context by connecting the content through procedurally generated objectives. But also to further develop EDD by implementing the artefact as a sub-system within the tool.
The Evolutionary Dungeon Designer (EDD) is a mixed-initiative design tool used for creating and generating dungeons [6]. Mixed initiative is a computer-human interaction approach within PCG where either the computer or the human can take the initiative and decide what to do next [6]. With this tool the designer can easily create dungeons by creating rooms of different sizes and connect them to each other using doorways. Each room can then easily be designed and modified tile by tile. The problem is that currently, the dungeons created with EDD lack context, and as stated above, generated content needs context to make
sense of the content [5]. So, the problem is less one of content generation than one of context building.
To address the challenge of context for procedurally generated content this thesis’s intention was to implement a sub-system for EDD which procedurally generates objectives for the dungeons created with the tool. The sub-system consists of macro patterns which are used to generate the objectives for each room. A macro pattern is an objective based on the existing content within the room. The macro patterns are based on the content in EDD which consists of spatial and inventorial micro patterns, as well as meso patterns, which are a combination of both types of micro patterns. By then using the created macro patterns this paper intends to answer the question:
How would a sub-system developed for EDD help create context in the form of objectives to represent the content in the generated dungeons?
This thesis follows the guidelines of the Design Science Research Methodology (DSRM), producing an artefact that procedurally generates these objectives. The artefact’s utility, quality and efficacy were then evaluated based on how well the objectives represent the layout of the dungeon with the help of experimental simulations. This sub-system acts as a way of automating the generation of objectives that will itself act as a narrative for the player to progress throughout the generated dungeon.
2. Related work
.2.1 Evolutionary Dungeon Designer
EDD is a tool for designers to create or generate dungeons [6]. With the tool the designer can create a dungeon by designing a set of rooms of various sizes that are connected to each other. The process of creating a dungeon is quite simple, when a room is created, it can be selected to be designed. These rooms are tile based, and each tile in a room can be modified to represent different types of paths, obstacles or rewards. The designer can e.g. place the player in the room, add doorways that can be connected to other rooms, place enemies and treasures. When the design of a room is complete, EDD then generates room templates through a genetic algorithm that base its calculations on what the handmade room looks like [7]. These can be selected at any time to replace the current design with the generated template to create unique room designs that match the designers style. The goal of this generation is to foster the designers creativity.
Figure 1: The main components in EDD. (a) A basic room, (b) different placeable tiles, (c) micro patterns and (d) meso patterns [9].
Each tile represents a piece of information and is used to create micro patterns. A pre-existing pattern finder in EDD analyses the tile information within a room to generate micro patterns which can be spatial or inventorial. Examples of micro patterns are chambers, corridors and connectors which are spatial, while bosses, enemies and treasure are inventorial, seefigure 1. Thus, when the micro patterns have been generated they are used to create meso patterns. Examples of meso patterns are guard rooms, treasure rooms and ambushes, see figure 1. Meso patterns, unlike micro patterns, are neither spatial nor inventorial. Meso patterns are composite patterns which are a combination of both. An example of this are the guard rooms which contain the spatial micro pattern of a chamber and the inventorial micro patterns of enemies [9].
2.2 Narrative
Jenkins, H. discusses the statement of Ernest, A. that interactivity and narrative are almost opposite of each other. With narrative, the author decides the direction of the flow, while interactivity turns to the player for motive power. Because of this, stories and games have conflicting demands. It is likely that straying from the author's path will make for a less satisfying story, but restricting the player’s freedom of actions will have the same effect on the game. But game designers are not simply story tellers, they sculpt and design game worlds and spaces. And not all games tell stories, there are more abstract games like Tetris, but games can never be reduced to simply experiencing a story, games are spaces with a possibility for narrative [10]. Generated content needs narrative in game space since it creates context which gives the content meaning. A lack of context in the generated content may result in a negatively affected user experience. Thus, making the content end up being perceived as empty or meaningless [5]. In From Hunt the Wumpus to EverQuest: Introduction to Quest Theory , Aarseth explores how landscape and quest types structure the gameplay and affect the story. The study also describes how the limitations of the story are based on the quest combinations available. If we can understand the structure of quests we can also understand the limits and potential of these kinds of games and how to create rich, open game worlds and tell interesting stories within them [11]. The common factor with objectives in games is to provide the player a reason to further progress through the game [5], something which EDD lacks at this moment. When generating content for game space; narrative or context, needs to be generated as well. Dormans et. al. investigates strategies to generate levels for action-adventure games. For this genre, level design is essential, and when procedurally generating levels for these games it is best to break down the generation process in two steps, one for generating game space and one for generating missions [8].
2.3 Procedural Narrative Generation
Ashmore et. al’s take on a generative system which procedurally generates narrative for their gameCharbitat, is an example of how an implementation of this results in a positive response from their tests. By basing the narrative generation on sets of tiles, which partition the game space, they create a graph which keeps track of the position of the player and also which new possibility of an objective is best suited to generate. This evaluation takes in mind previous objectives, and actions, done by the player. Thus, resulting in a more personal experience while also increasing the replayability [5].
Kybartas et. al’s survey on generative narrative techniques explores the development of different techniques used in not only the game development industry, but also in academic settings. When discussing this subject, the problem is divided into two tasks which both need to be created: plot and space, either automatically or manually generated. Plot is defined as a set of events with an overall structure which represents both the temporal ordering, and the causal relations between the events. Space includes the characters, settings, props and anything which is present either physically or abstractly in the space of the narrative [12]. The survey explores a plethora of previous work done in the area of procedural narrative generation and concludes that one artefact called Slant, developed by Montfort et. al [13], is an example of a well implemented version of narrative generation. Dormans et. al [8] and Kybartas et. al [12] both claim that to generate narrative requires the two aspects mentioned above: plot, and space. By generating space they also generate context for it, thus creating a unique narrative for each possible outcome of the generative process.
3. Methodology
Hevner et. al [14] discusses two paradigms which characterize much of the research in the Information Systems discipline. These are the Design Science Research Methodology (DSRM) and behavioral science, where we have chosen the former to be our methodology for this thesis. By focusing on creation and evaluation of artefacts that have the goal to solve identified problems, it became natural to adapt the DSRM since its basis is to inherently be a problem-solving process through the development of technological artefacts. Hevner et. al. presents seven specific guidelines of evaluation and iteration within research projects, which shall be followed throughout the development of our artefact. The guidelines are as follows:
1. Design-science research must produce a viable artefact in the form of a construct, a model, a method, or an instantiation.
2. The objective of design-science research is to develop technology-based solutions to important and relevant business problems.
3. The utility, quality, and efficacy of a design artefact must be rigorously demonstrated via well-executed evaluation methods.
4. Effective design-science research must provide clear and verifiable contributions in the areas of the design artefact, design foundations, and / or design methodologies. 5. Design-science research relies upon the application of rigorous methods in both the
construction and evaluation of the design artefact.
6. The search for an effective artefact requires utilizing available means to reach desired ends while satisfying laws in the problem environment.
7. Design-science research must be presented effectively both to technology-oriented as well as management-oriented audiences.
Since this thesis is directly connected to the development of an IT artefact that tries to provide a solution to a business problem, it felt that the DSRM was a suitable research methodology. One of the strengths of the DSRM is that it is an iterative process, which helps in prototyping different iterations of the IT artefact in order to find the best solution under the set restrictions such as time, money and scope [15].
3.1 Guidelines of evaluation
. 3.1.1 StructureProducing a viable artefact for the problem of creating narrative for procedurally generated content can be made in several different ways. Some examples are either implementing the tools necessary for the player to design their own objectives in the dungeons, generating a set of objectives the designer can place in the dungeon and letting a system design the rooms based on the objectives, or completely procedurally generating the narrative based on the design of the dungeon. We have developed the latter artefact for EDD as a sub-system that generates objectives for the player to complete, creating context for the player. This system takes the generated dungeon and creates macro patterns for rooms based on the meso patterns that exist in them. As mentioned in the 1. Introduction, a macro pattern is an objective based on the existing content within the room, and content consists of micro and meso patterns. The in depth definition of the macro pattern can be found in 4.2 Macro patterns.
3.1.2 Relevance
Within the game development industry, PCG has shown to be a reliable method to decrease the workload and therefore decrease costs. This has been done to both mass-generate content, but also using it as a mixed-initiative tool which can foster creativity for the design team [1, 2, 3]. Thus, making PCG a natural method for both smaller and larger development teams to find a solution to such problems which would otherwise be too large, costly or time consuming. As mentioned in 1. Introduction, a challenge with this is a lack of context when implementing the procedurally generated content into the game space. This thesis addresses this challenge by developing a technological artefact that generates the necessary context needed to make sense of the content produced with EDD to expand on the knowledge of procedural narrative and its generation.
3.1.3 Evaluation
To evaluate the artefact, an experimental evaluation has been conducted through simulation. The simulation has a quantitative approach and the goal of the simulation is to make sure that the system creates appropriate objectives based on the layout of the dungeon in every type of scenario. The scenarios are short and long dungeon layouts, those with a single path through the dungeon, ones with several dead ends, circular layouts as well as a mixture of both. The simulation is conducted through designing a dungeon layout and evaluating the main and side objectives that the developed system determines fit for the dungeon. The result of the evaluation will be based on how well the system generates objectives for the layout of the dungeon, or how well it adapts and uses the game space. The results should reflect the intentions of the designer when creating the dungeon. Furthermore, evaluating the system’s reaction to a changing layout is crucial to be able to create results that make sense for the dungeon.
3.1.4 Contributions
This simulation contributes to EDD by presenting a solution to implementing procedural narrative within game related content. The simulation results show the effectiveness of our approach by determining the quality of the artefact through comparison of the produced results and the dungeon layout. The results can be used as inspiration or knowledge when trying to expand or improve on the solution or when working within the field. As mentioned in 2.2 Narrative, narrative in games have a strong connection to the game space. It is
designed to represent the narrative in the game and to create immersion and atmosphere in the same way as traditional media such as books and films without restricting the player from freedom of action [10]. Because of this, we value the usage of game space very highly in our evaluation of the design artefact. What this means is that the layout of the dungeon will be the determining factor in the evaluation of the artefact, because we do not want parts of the dungeon to feel empty or meaningless.
3.1.5 Research Rigor
The research will address the lack of contextuality in the current iteration of EDD, the need of the artefact for game designers, and focusing on laying the grounds for future iteration of objective generation. The artefact will have a theoretical base through previous research about the importance of contextuality, narrative, and procedural quest generation. Also, by following research methodologies and experimentation methods. The artefact will be iteratively tested, covering possible scenarios of layouts designed in EDD.
3.1.6 Effective artefact
To produce an effective artefact and reach the desired results of creating procedural narrative for the dungeons designed with EDD, the artefact will utilize the already existing base for patterns, views, rooms and dungeons and build upon them to implement objectives in the dungeons and generating their narrative. The development of the macro patterns will extend on the meso patterns to generate the objectives of the rooms since the existing patterns represent the content within the room. As for displaying the objectives, or narrative, the artefact will make use of the world view, where the dungeon is shown, and display the objective icons over their assigned rooms to clearly indicate the type of objective and show which objective is the main objective through a differentiating colour.
3.1.7 Presentation
The simulation is conducted in a way to gather results on how well the procedural narrative is implemented. The result will be based on the objective quality measurement which is backed up by the research of game design as a narrative architecture [10]. The presentation of the results will be screenshots that showcase all the different types of objectives, which are, defeat a group of enemies, a boss or find the treasure room, in various dungeon layouts, e.g. a dungeon with a single path, several dead ends or a circular layout. This will visualise to the reader how well the system adapts to the layout, that it can create procedurally generated objectives for the player in each scenario and in which areas the system can be improved. In this way, the reader can understand the thesis, its goals and the results of the artefact regardless of their technological background. Then, by also describing the artefact regarding its development, structure and functionality, a base of knowledge for further development and improvements of the research is created. Thus, opening up for repeatability and advancements for this thesis’ subject.
3.2 Method discussion
The chosen method to test our artefact is through simulation. Testing is a crucial part of any iterative process to be able to achieve the best results. Therefore, the method of simulation suits the needs of testing, evaluating and iterating for our artefact. The plan of testing was to start by implementing the macro patterns in the tool, the evaluation methods for the dungeon and its objectives as well as the narrative filter that displays them in the dungeon. When a basic implementation of the system was ready for testing, different dungeon layouts would be
designed with the tool to test the evaluation of the objectives implemented by the system and then adjust it to make the system reactive and adaptable with all the different layouts. When evaluating an artefact like this, there are several methods of testing the artefact and its quality. One other method of evaluation would be to conduct a user study and either have designers create dungeons and see how well the system reacts and adapts to the expected results of the designer, or to have users play through dungeons designed with the tool and have them evaluate the objectives for the dungeon layouts. Combining an iterative testing process with a user study could be an effective evaluation method, though with our timeframe it would decrease the scope of our evaluation method. Furthermore, since EDD does not, at this point in time, provide a playable version of its dungeons we realised that a user study would not be a suitable way of testing and evaluating our artefact.
4. Artefact
.4.1 Design & Technology
In design terms the artefact is a visual filter that the designer can use to show the objectives procedurally generated for a dungeon created with the tool. The visual filter can be toggled on and off with a button from the world view, the button can be seen in figure 2. When the objectives are toggled on they are shown as an icon on top of the room assigned with the objective, as well as with a coloured background covering the room. The icon represents the type of objective and the colour of the background indicates if the objective is the main objective or one of the side objectives in the dungeon, see figure 3.
Figure 2: A part of the button pane in the world view. (a) Move rooms, (b) connect rooms, (c) move player, (d) start with suggestions and (e) toggle objectives.
Figure 3: Every type of dungeon objective as both main objective (green) and side objective (blue). The icons represent the different types of objectives (a) “Defeat the enemies”, (b) “Find the treasure” and (c) “Defeat the boss”.
In technical terms the artefact is an objective generator for EDD that uses the layout and the content of a dungeon for its generation and evaluation process. The generated objectives that the artefact produces are based on patterns, and these different patterns represent content in a
room of the dungeon, and each room has their own patterns and objective. The artefact then decides the amount of dungeon objectives based on the size of the dungeon, and the importance of the room objective based on the room location in the dungeon layout. The usage of game space is the determining factor in the evaluation of the dungeon objectives. When the dungeon objectives have been assigned, the dungeon signals the renderer which rooms shall display their objectives when the button is pressed. The objectives for each dungeon are also deterministic, meaning that the same dungeon design will always result in the same objectives.
4.2 Macro patterns
In the artefact, macro patterns represent the objectives in the dungeon. However, to explain what a macro pattern is, we first have to explain what micro and meso patterns are. There are two types of micro patterns, spatial and inventorial. A spatial micro pattern represents a type of area, for example a chamber or a corridor, and an inventorial micro pattern represents a type of content, usually tile sized, for example an enemy or a treasure. Inventorial patterns are then assigned to spatial patterns, meaning that a spatial micro pattern contains the inventorial micro patterns in its area. These micro patterns are then used to create meso patterns. A meso pattern is a composite pattern, which is a combination of both spatial and inventorial patterns. Together the spatial and inventorial patterns create points of interest in rooms, for example treasure chambers, guard rooms or ambushes which are stored in composite patterns as meso patterns. These meso patterns, or points of interest, are used to create the macro patterns, which are a selection of composite patterns, and each macro pattern uses different types of meso patterns for their objective. The macro patterns look for all relevant meso patterns in the room and use the one with highest quality for the objective. For example, in a room with several guard rooms only one macro pattern would be created, and it would represent the guard room with the highest quality. Furthermore, the quality of the several guard rooms in the rooms does not add up to increase the quality of the macro pattern. In another example, where there are several meso patterns in the room, but of different types, one macro pattern of each type would be created and the quality of the macro patterns would be compared to select the best one for the room objective. In this iteration of the artefact there are three different macro patterns: finding the treasure, defeating the enemies and defeating the boss, seefigure 3.
4.3 User Interface
For this thesis, there are two main views in EDD the reader needs to be familiar with, these are the WorldView and the RoomView.
Figure 4: To the left (a) is the WorldView or the overview of the dungeon. In this view, the user can navigate the overview of the dungeon and change the layout of the dungeon with the help of the buttons on the right-side
panel. To the right (b) is the RoomView or the edit view of a room. In this view, the user can change the design of the room with the tiles on the left side of the view or the room design suggestions on the right.
The WorldView displays the entire dungeon and an overview of each room and how they are connected to each other. In this view, the designer has a few options as well. The designer can add grid-based rooms of size MxN to the dungeon, as well as remove any of the existing rooms. The designer can also use five different buttons to move rooms, connect rooms to each other, place the player on any tile in any room of the dungeon, start with a suggestion and toggle on and off objectives, see figure 2 and (a) in figure 4.
The RoomView displays the room that has been selected for editing. In the RoomView the player can design the selected room by modifying any tile as floor, wall, treasure, enemy or boss with the buttons on the left side of the user interface, see (b) in figure 4. Just below the various tile buttons the designer can press another button to display the existing micro and meso patterns in the room, which are used for the generation of the objectives in the dungeon. On the right side of the user interface the tool is displaying a set of genetically generated room suggestions which can be selected at any time to replace the current design of the room, see (b) in figure 4.
The user interface for the objectives that is generated with the artefact we have developed adds a button in the WorldView, that the designer can use to toggle on and off a filter which displays the main and side objectives in the dungeon they have created, see figure 2 and (a) in figure 4 . These objectives will be displayed similarly to how the patterns are displayed in the room view, showing icons over the rooms with objectives in them and with a coloured background that indicates if the objective is the main objective or one of the side objectives.
4.4 Flow of the app
When the user has created a dungeon and designed its rooms, they can press the toggle objectives button to show the dungeon objectives, see figure 2. When the button is pressed each room in the dungeon finds its micro and meso patterns, or generates them if necessary, to then generate its macro patterns. The rooms are then assigned an objective based on the macro pattern with the highest quality in that specific room. When all rooms have been assigned an objective, the dungeon calculates the amount of objectives needed for the dungeon, see equation 1 in section 4.5 Metrics. It then evaluates the importance of each room objective based on the dungeon layout and the room objective quality. The objective with the highest priority is set to the main objective of the dungeon. The amount of objectives needed is stored in the dungeon and the remaining room objectives are discarded. The rooms with an assigned dungeon objective is then signaled to render an icon on top of the room representing the objective of the room as well as a coloured background that indicates if the room contains the main objective or one of the side objectives.
4.5 Metrics
To get the most accurate and best results from the artefact we are developing, it is important to make sure that the correct metrics are used when evaluating the generated objectives. Therefore, when evaluating the objectives, the following metrics have been chosen:
1. The distance between the player and the objective. Having to interact with the content is an important factor, e.g. completing all the objectives before exploring the dungeon from start to finish would break the players’ immersion [10].
2. If the room is a dead end. Unless the designer creates an empty room, there is most likely a purpose of the room's existence. And if the room is a dead end, there is a higher chance of the player expecting the path to take them somewhere [5]. Therefore, to not make the outskirts of the dungeon feel meaningless, they will always be assigned an objective.
3. The amount of connections to other rooms. If a room is connected to many other rooms in the dungeon, the chance of the player stumbling upon the room is higher than one with less connections. Therefore, to create an initiative to explore these rooms, the rooms with less connections to other parts of the dungeon will be prioritised to have an objective.
4. The evaluated quality of the room. As well as the importance of the dungeon layout and how well it is used for the placement of the different objectives, another important metric is the quality of the objective itself, which it inherits from the meso pattern.
5. The amount of objectives in the dungeon. It is important to calculate a suitable amount of objectives for a dungeon to not make the dungeon feel empty, meaningless or crowded with an abundance of objectives. Therefore, the amount of objectives is calculated with an equation that tries to utilize the game space, see equation 1.
Equation 1: Used to calculate the amount of objectives in a dungeon. “de” equals the amount dead ends and “r” equals to the amount of rooms
In addition, if a dead end is the initial room, which means it contains the player, the dead end is excluded, but the room is still used for the rest of the equation. The remaining rooms are divided by four because it resulted in a suitable amount of objectives for the dungeon designs tested, but the equation can be improved to be more adaptive to the design of the dungeon. So, if we use the dungeon layout from the simulations in figure 6 as an example, the amount of objectives becomes two, because there is one dead end, excluding the initial room as a dead end, and a remaining four rooms which divided by four adds another objective.
5. Results
Our results consist of a total of 14 simulations in EDD with different layouts, room sizes and content. The results will show how the artefact reacts and adapts to the different scenarios when evaluating the dungeon objectives.
The first simulation displayed in figure 5 shows a dungeon with (a) the initial room of the dungeon with the player and (b) another room with the main objective of “Defeat the boss” based on the information provided in figure 3. The goal of the simulation is to show that the system is capable of creating objectives in a dungeon with the most simplistic dungeon layout and giving a brief introduction to how the dungeons and system of the artefact work.
5.1 Small dungeon layout simulations
Figure 6: A small single path dungeon layout with a main objective (green) and a side objective (blue) in two different scenarios.
In the two simulations shown above in figure 6 we are testing the system in a rather small dungeon layout with all rooms connected sequentially through the dungeon. The purpose of the two simulations in figure 6 is to show how the system reacts and adapts to the change in the dungeon layout to maximize the usage of game space when generating its objectives. In simulation (a) the initial room is to the far left and the room on the far right becomes the main objective, using all of the game space, and the room next to the main objective becomes the side objective based on the rooms location in the layout and distance from the initial room. However, in simulation (b) the initial room is moved and the system reacts and adapts to this change by changing the side objective to the left side dead end to continue utilizing the game space of the dungeon layout.
In figure 7 above we are testing a small circular dungeon layout that doesn’t have any dead ends. The goal of these simulations are to test the systems capabilities to also generate suitable objectives for dungeon layouts without dead ends. In this layout there is only one objective because of the lack of dead ends and a total of four rooms. To utilize most of the dungeon layout in this scenario, the room on the opposing side of the initial room in the dungeon gets assigned with the main objective. That is because of its distance from the initial room, regardless of the type, or quality, of the objectives in the rooms.
5.2 Large dungeon layout simulations
Figure 8: A large dungeon layout with several dead ends, a main objective (green) and four side objectives (blue) in two different scenarios.
In figure 8 we conduct simulations to test the systems capabilities of generating suitable objectives for larger dungeon layouts with several dead ends, extending on the simulations in figure 6 . In simulation (a) the initial room is in the middle of the dungeon layout and every dead end is assigned either as a main or a side objective to utilize the game space fully. One interesting thing to bring up about this dungeon layout is that the two “Find the treasure” objectives are both dead ends and at the same distance from the initial room, see the red bordered rooms in figure 8. So to determine which of the objectives should be the main objective, the system compares the objective quality of the two rooms. The first of the two objectives that is evaluated based on the dungeon layout will be assigned the main objective, and when the other objective is evaluated it has to have a higher quality to replace the current main objective. The same evaluation is made for the side objectives. In simulation (b) the initial room has been moved and now the main objective has changed, but the previous main objective still remains an objective to utilize as much of the game space as possible.
Figure 9: Two large circular dungeon layouts without dead ends, a main objective (green) and (a) one side objective or (b) three side objectives.
The two different dungeon layouts from the simulations of figure 9 is an extension of the dungeon layout from the simulations in figure 7. The goal of simulation (a) in figure 9 is to observe if the system can generate objectives for a larger circular dungeon layout where the corner rooms of the dungeon have no objectives in them. As can be seen in the simulation, the system, because of the lack of content in the corner rooms of the dungeon layout, instead assigns the rooms furthest from the initial room as the objectives to utilize as much game space as possible. Similar to simulation (a) in figure 8 the two objectives in the dungeon are equally distant from the initial room and the system evaluates that the “Defeat the boss” objective has the highest priority based on the quality of the objectives.
In simulation (b) the corners of the dungeon layout contain content, and the system utilizes this to generate objectives in every corner of the dungeon to maximize the usage of game space. The remaining non-corner “Find the treasure” side objective is prioritized over the other rooms without an objective based on its distance from the initial room.
5.3 Combined dungeon layout simulations
Figure 10: A large circular dungeon layout with dead ends, a main objective (green) and four side objectives (blue) in two different scenarios.
Figure 10 introduces a large circular dungeon layout with dead ends. This layout is a combination of previous simulation dungeon layouts to show that the system not only can react and adapt to both types of layouts but a combination of them as well. In simulation (a) the initial room is part of the circular center of the dungeon layout and the system generates objectives in the various dead ends of the layout to utilize the game space. In addition, there is a final side objective of type “Defeat the boss” which is prioritized because of both its distance from the initial room and since it has less connections to other rooms in the dungeon in relation to the non-objective guard rooms.
In simulation (b) the system reacts and adapts to the change of initial room and another treasure room becomes the new main objective based on its distance from the initial room. However, the system now generates an objective in the circular structure of the dungeon layout. And as in previous simulations in figure 7 and figure 9, the system prioritizes the room in the farthest corner from the initial room of the circular layout with content within it for its dungeon objectives. Furthermore, “Defeat the boss” is still a side objective since it is connected to less rooms than the remaining non-objective treasure and guard room, meaning that it has a lower chance of being interacted with by chance and in general needs a higher priority to become an objective.
5.4 Dungeon content simulations
Figure 11: A large circular dungeon layout with dead ends, rooms of different sizes and varying content, a main objective (green) and three side objectives (blue) in three different scenarios.
Figure 12: Two different room designs shown with and without meso patterns toggled. (a) and (c) as well as (b) and (d) are the same design. The two room designs represent the bottom room from the figure 11 dungeon layout.
In the final simulations displayed in figure 11 the dungeon layout is a variation of the layout in figure 10. In the previous simulations the goal has been to show how the system reacts and adapts to different types and sizes of dungeon layouts with minor changes to the dungeon content. In these simulations in figure 11, the goal is to show how the change of dungeon content affects the system and its evaluation of the objectives.
In simulation(a) we changed the sizes of the different rooms in the dungeon to show that the system does not include the size of the room in the evaluation of the room objectives. Therefore, the objectives in simulation (a) in figure 11 are near identical to simulation (b) in figure 10. The only difference with the objectives is that the “Defeat the enemies” objective in the circular structure of the dungeon is removed, see the red bordered rooms in simulation (b) in figure 10 and simulation (a) in figure 11. The reason is because the two middle rooms in the circular structure are combined into one big room, reducing the amount of objectives in the dungeon layout.
In simulation(b) we change the content of the dungeon layout once again. This time, the big room in the middle of the dungeon layout is filled with several meso patterns to show that the system does not prioritize the amount of content in a room when evaluating the dungeon objectives. The usage of game space in the dungeon layout is the top priority when the system generates its objectives. In the case of the room becoming the main or a side objective, the macro pattern with the highest quality in the room would become its objective. In simulation(c) we showcase the scenario of a room having several macro patterns and how the system generates its objective. The goal of the final simulation is to show that the system generates objectives based on the content in the dungeon, meaning that even if the system
procedurally generates the objectives with the focus on using the game space, the designer of the dungeon is still in control of what objectives are created by designing the layout and content of the dungeon. The difference between simulation (b) and (c) in figure 11 is the side objective in the bottom room of the dungeon. In figure 12 we can see two different room designs, with and without meso patterns. (a) and (c) in figure 12 represent simulation (b) in figure 11 and (b) and (d) represent simulation (c) in figure 11. In both room designs there are two treasure chambers protected by an ambush, but changing the contents of the room will in turn affect the quality of different objectives in the room and a new room objective may be selected as can be seen in simulation (b) and (c) in figure 11.
6. Discussion
When the testing of our artefact was conducted, the utmost importance was to test the flexibility of our system and see if the system was capable of generating objectives, or context, for the content created in EDD with the help of macro patterns. The figures shown in 5. Results is our way of testing all various types of layout scenarios that could occur when a designer uses the tool. Therefore, we consider that the simulation results successfully showcase our intent to answer our research question. Moreover, the results also allow us to come to a conclusion since the artefact is capable of procedurally generating suitable objectives for the various dungeons designs from our simulations, using the macro patterns as a base for the objectives, and successfully creating context in the dungeons.
However, while our simulations have extensively tested the artefact and different dungeon layouts and designs, we can not prove that it will give the best results in every possible dungeon design. The results from the tests of the artefact allow us to argue that the artefact successfully uses most, if not all, of the game space available in the dungeon layouts with minor downsides. But there is a difference in the quality of the results based on the types of dungeon layout, referring to the usage of game space. When testing dungeon layouts with dead ends the system generally uses more of the game space than in circular layouts. This becomes more apparent when the two types of layouts are combined for the later simulations, see figure 10. The system favours the dead ends to utilize the game space and objectives are only present in the central rooms of the dungeon when all dead ends have been assigned an objective and the amount of dungeon objectives are fewer than the amount of dead ends in the dungeon design. Therefore, you can argue that the center of the dungeon feels more empty because of this. But from simulations in figure 7 and 9 we see the same type of results even without dead ends, meaning that the system might have a harder time utilizing game space in layouts with several paths to the same objective.
An interesting topic for the discussion is the choice of making the algorithm for the dungeon objectives deterministic or stochastic. In the current state of the artefact, the algorithm is deterministic which means that the same dungeon design will always produce the same dungeon objectives. The approach was chosen since we believe it was most suitable for our testing purposes. An alternative approach to this is to generate stochastic dungeon objectives, meaning that the system will generate different objectives for the same dungeon design every time. Though, these two approaches of generating objectives for dungeon designs have different purposes. A deterministic approach gives the designer more control over the objectives generated for the content produced, though at the cost of a variety of objectives for the dungeon designs. However, the stochastic approach is the polar opposite. With this approach the system can generate endless combinations of objectives for the same dungeon
design which creates more variety, but this leads to the designer losing control over the final result of the content created with the tool. Therefore, the different approaches are most suitable for different purposes and scenarios based on the needs of the designers.
In Orchestrating Game Generation [16], Liapis et. al. discusses the necessity of a harmonizing communication across domains, for instance within a game. A great example of this would be a system which could procedurally model and animate a character, which would generate content that is meaningfully related and intertwined, and not just a layered creation. This would achieve the harmonization, where the two parts of the system orchestrates the various computational creators. Dormans et. al. talks about dividing the generation process in two steps, one for the game space and the other for context [8]. So, by dividing the generation process within EDD, we followed in Dormans’ footprints to not lose the existing scope of the system. This integration between the two systems gives the designer the opportunity to view what objectives the designed layout will have and from there on redesign that layout to better fit what the designer has in mind for the context of the game space. Thus, you can argue that the artefact which has been developed and EDD, being our two creative domains, do work together in a harmonious way.
7. Conclusion
When procedurally generating content, there is a need for context building or the content may end up being perceived as empty or meaningless. By dividing the generation process into two parts, game space and context, we were able to develop an artefact that can react and adapt to various different dungeon designs created with EDD and generate suitable objectives based on the content in the dungeon and utilize the game space. With the results from our experimental simulation tests, macro patterns have proven to be a viable solution to the problem of context building for procedural content generation in the Evolutionary Dungeon Designer. By developing this artefact we have contributed knowledge to the area of procedural narrative generation by presenting a viable solution for the implementation of context in content generation tools. The developed artefact is also a sought after extension for EDD to help designers get a better visual understanding of the content they are creating. Thus, giving the possibility of further fostering creativity in a different aspect from what was previously possible within EDD. Moreover, the developed artefact tries to alleviate parts of the workload from the designer regarding the narrative of the dungeon, which both end up as a way of saving time but also as a tool for the designer to validate that their creation is designed as they had in mind in terms of narrative.
8. Future work
By continuing the work presented in this paper, further improvements and iterations can be tested and implemented, for example:
1. Expanding the evaluation’s depth, increasing the input of variables to further grant a higher level of detail in the objectives created. For example; by using neighbouring rooms’ macro patterns, creating a chain of objectives that together can be held to a higher value than in this iteration of the artefact. Furthermore, with this expansion of the evaluation process, one could also add more macro patterns to diversify the outcome of the objectives. Thus, creating not just objectives for the player to complete, but rather a coherent goal throughout the dungeon. Moreover, all objectives start from the initial room or player position. A suitable extension of the artefact
would be to implement “quest givers” or non-playable characters that distribute the objectives of the dungeon. This would change their start position which could utilize more game space and allow for more varied objective combinations and dungeon narratives.
2. Conduct a user study to gather data on how the objectives generated compare to what a designer, or player, expect should be the objectives of certain dungeon layouts, and how the objectives can affect the replayability of the dungeon. Though this would be more suitable for a future iteration of the artefact that has had further advancements in the code, such as the previously mentioned example.
References
[1] Shaker, N., Togelius, J., & Nelson, M, J. “Procedural content generation in games". Switzerland: Springer International Publishing, 2016.
[2] Hastings, Erin J., Guha, R. K., & Stanley, K, O. "Evolving content in the galactic arms race video game".
2009 IEEE Symposium on Computational Intelligence and Games. IEEE, 2009.
[3]Alvarez, A., Dahlskog, S., Font, J., Holmberg, J., Nolasco, C., & Österman, A. "Fostering creativity in the mixed-initiative evolutionary dungeon designer." Proceedings of the 13th International Conference on the Foundations of Digital Games. 2018.
[4] Togelius, J. Champandard, A. J., Lanzi, P. L., Mateas, M., Paiva, A., Preuss, M., & Stanley, K. O.
“Procedural content generation: Goals, challenges and actionable steps.” Schloss Dagstuhl-Leibniz-Zentrum fuer Informatik, 2013.
[5] Ashmore, C, & Nitsche, M. "The Quest in a Generated World." DiGRA Conference. 2007.
[6]Baldwin, A., Dahlskog, S., Font, J. M., & Holmberg, J. "Mixed-initiative procedural generation of dungeons using game design patterns." 2017 IEEE Conference on Computational Intelligence and Games (CIG). IEEE,
2017.
[7] Alvarez, A., Dahlskog, S., Font, J., Holmberg, J., & Johansson, S. "Interactive Constrained MAP-Elites Analysis and Evaluation of the Expressiveness of the Feature Dimensions." arXiv preprint arXiv:2003.03377 (2020).
[8] Dormans, J., & Bakkes, S. "Generating missions and spaces for adaptable play experiences." IEEE
Transactions on Computational Intelligence and AI in Games 3.3 (2011): 216-228.
[9] Alvarez, A., Dahlskog, S., Font, J., Holmberg, J., & Johansson, S. "Assessing aesthetic criteria in the evolutionary dungeon designer." Proceedings of the 13th International Conference on the Foundations of Digital Games. 2018.
[10] Jenkins, H. "Game design as narrative." Computer 44.53 (2004): 118-130.
[11]Aarseth, E. "From hunt the wumpus to everquest: introduction to quest theory." International Conference
on Entertainment Computing. Springer, Berlin, Heidelberg, 2005.
[12] Kybartas, B., & Bidarra, R. "A survey on story generation techniques for authoring computational narratives." IEEE Transactions on Computational Intelligence and AI in Games 9.3 (2016): 239-253.
[13]Montfort, N., y Pérez, R. P., Harrell, D. F., & Campana, A. "Slant: A Blackboard System to Generate Plot, Figuration, and Narrative Discourse Aspects of Stories." ICCC. 2013.
[14]Hevner, A., R., March, S. T., Park, J., & Ram, S.. "Design science in information systems research." MIS
[15]McDonnell, J. "Gifts to the future: Design reasoning, design research, and critical design practitioners." She
Ji: The Journal of Design, Economics, and Innovation 1.2 (2015): 107-117.
[16] Liapis, A, Yannakakis, G. N., Nelson, M. J., Preuss, M., & Bidarra, R. "Orchestrating game generation."
IEEE Transactions on Games 11.1 (2018): 48-68.
References (Games)
Minecraft, Mojang, 2009
No Man’s Sky, Bourne G. for Hello Games, 2016
Spelunky, Mossmouth LLC, 2008
